The invention discloses a distributed multi-sensor fusion algorithm based on AMDs. The algorithm comprises the following steps: S1, initializing system parameters; S2, using a constant false alarm detector to process radar echo information to obtain a measurement information set; S3, independently estimating a target state based on a kalman filter, S4, using high dimensional Gaussian distribution to represent the multi-objective joint posterior probability density; S5, carrying out the dimensionality reduction operation on the joint posterior probability density; S6, sending the AMD of each node to the adjacent nodes; S7, fusing the AMDs by using a generalized cross-covariance algorithm; S8, extracting the target state; S9, setting k = k + 1, if k>K, outputting the target state extracted in the S8 as a track; otherwise, returning to step S2. The distributed multi-sensor fusion algorithm realizes the joint fusion of a plurality of targets under the condition that the estimation errors of the different sensors are considered cross-correlated, has higher adaptability and better robustness, and effectively solves the problem of multi-target joint posteriori fusion in the traditional tracking system.